>Tcsh does this is you let it use its own malloc, IIRC.   I'm not sure
>whether it's a bug in their malloc or a bug elsewhere in tcsh that our
>malloc doesn't catch, but when I recompiled tcsh using our malloc, it
>didn't print the message and hasn't dumped core on me.   That's my
>recollection, anyway - it's been more than six months since I compiled
>tcsh for the pmax, so I may be forgetting something.

I build tcsh static and install in /bin,, because I use it as root's
login shell. That works fine.


There seem to be consistent problems with linking in an alternate
malloc() implementation unless you link statically. Ive seen similar
buggy behaviour with xterm's resize command.  Either using the
standard BSD malloc (i.e., dont link against libgnumalloc), or linking
statically, fixes the problem. TThere's also a very similar problem
with perl5, if/when it links in the POSIX-ized version of rename(2)
as well as the rename(2) in libc: the resulting perl coredumps
when testing rename.

I strongly suspect a bug somewhere in eithe rthe linker or the dynamic
linker (ld.so).  I don't know who to tell which it is, though.